Fusion in medical imaging: theory, interests and industrial applications.

The accumulation of several data coming from medical images and signals, expert knowledge and databases is becoming very common for the study of a given pathology. The aggregation of all this information is mentally performed by a clinician, and generally allows for a better medical decision in clinical studies. We propose in this article a fusion method that models this aggregation process. This method is a three step scheme, that first transforms all the available information in a common theoretical frame, then aggregates these data using their redundancy and their conflicts, and finally computes a new information synthesizing all the initial knowledge. We first introduce in this article the fusion scheme and its theoretical aspects, and we particularly focus on the three steps of the process. We then detail the software implementation of this concept, achieved in collaboration with SEGAMI Corporation Inc. We finally apply this concept to a real clinical problem, the study of Alzheimer's disease using MR and SPECT images, and we show very encouraging preliminary results.

Iterative reconstruction methods for nonuniform attenuation distribution in SPECT.

Two iterative methods, a generalization of the Chang method and a projection precorrection, were investigated to determine whether the use of an attenuation map could improve nonuniform attenuation compensation. After a detailed description of the methods, results obtained with simulated and phantom data were compared. This study demonstrates that projection precorrection provides accurate quantification and good image quality as early as the precorrection step, whereas the generalized Chang method requires computation of one more iteration.

The value of pixel per pixel multiharmonic Fourier analysis to assess left ventricular function.

Left ventricular systolic and diastolic function can be assessed by peak ejection and filling rates and their time of occurrence. These parameters can be calculated using two different methods: from the global left ventricular time-activity curve analysed with a four harmonic Fourier fit and from each pixel time-activity curve analysed with two and three harmonics (the values being averaged over the left ventricular region of interest). In both cases, values were normalized for heart rate and end diastolic counts. A study was conducted in a series of 11 patients (six without and five with a previous myocardial infarction but a normal left ventricular function) examined at baseline and during an i.v. dobutamine infusion, at a dose known to increase both peak ejection and filling rates. During dobutamine infusion, analysis of global left ventricular time-activity curve demonstrated a statistically significant increase in both peak ejection and filling rates, but the local analysis showed a more significant increase of these parameters. To assess ventricular function, a local harmonic analysis can be used and appears to be a more sensitive approach than analysis of the global left ventricular time-activity curve. The local analysis provides spatial mapping and a histogram of the parameters which can be used as parametric images to describe systolic and diastolic function.

Improved detection of anterior left ventricular aneurysm with multiharmonic fourier analysis.

Single and multiharmonic Fourier analysis of LAO 30-45 degrees gated blood-pool studies were performed in a selected group of 30 patients with a left ventricular anterior aneurysm proven by contrast angiography. The sensitivity of the first harmonic phase image for the diagnosis of ventricular aneurysm was 80%. The clear phase shift (greater than 110 degrees) between the normal and the aneurysmal areas was missing in six patients. Peak acceleration images (negative maximum of the second derivative of the Fourier series) were calculated for each pixel with the analytical Fourier formula using two or three harmonics. A clear phase shift (greater than 126 degrees) than appeared in all the patients. This improvement was related to the increased weight of the second and third harmonics in the aneurysmal area when compared to control patients or to patients with dilative cardiomyopathy. Multiharmonic Fourier analysis clearly improved the sensitivity of the diagnosis of anterior left ventricular aneurysm on LAO 30 degrees-45 degrees gated blood-pool images.

Filters and Fourier analysis of gated blood pool studies: a search for the optimal combination.

Fourier analysis of gated blood pool studies is performed after filtering the raw data by a spatial median 3 x 3, 9 x 9 or temporo-spatial 9 x 9 x 9 filter. 20 patients and a dynamic cardiac phantom were studied to determine the quantitative effects of these filters and of multiharmonic Fourier filtering (MHFF). The filtered MHFF data, with or without preprocessing, were compared with a 3 D or 2 D filter to the raw data using a chi 2 distribution. The MHFF (two or three harmonics) procedure applied to the raw data of patients without any preprocessing produced the smallest chi 2 value, thus demonstrating the very close relationship between filtered images and raw data. Preprocessing the raw data by the median filter also preserved the signal when two or three harmonics were applied, whereas the 3 D and 2 D (9 x 9) filters did not. The phantom study also demonstrated that MHFF preserved the signal better than any other preprocessing. The median filter introduced a smaller distortion than the 2 D (9 x 9) and 3 D filters. It is concluded that MHFF applied with two or three harmonics on the raw data or after preprocessing by a median (3 x 3) filter is the most successful way of preserving the real signal. It is believed that the other filters should be avoided. The clinical advantage of MHFF processing is to provide both very accurate filtering and parametric images.

Two-dimensional mapping of three-dimensional SPECT data: a preliminary step to the quantitation of thallium myocardial perfusion single photon emission tomography.

A method is presented by which tomographic myocardial perfusion data are prepared for quantitative analysis. The method is characterized by an interrogation of the original data, which results in a size and shape normalization. The method is analogous to the circumferential profile methods used in planar scintigraphy but requires a polar-to-cartesian transformation from three to two dimensions. As was the case in the planar situation, centering and reorientation are explicit. The degree of data reduction is evaluated by reconstructing "idealized" three-dimensional data from the two-dimensional sampling vectors. The method differs from previously described approaches by the absence in the resulting vector of a coordinate reflecting cartesian coordinate in the original data (slice number).

Interrogation and display of single photon emission tomography data as inherently volume data.

The purpose of single photon emission computed tomography (SPECT) data is to map the tracer concentration from the three-dimensional object into a three-dimensional image array. The conventional interrogation of the data is through slice interrogation. In this paper we explore display methods in which the data are directly interrogated and processed as three-dimensional data. This includes direct addressing of sagittal, transverse, and frontal slices, around a targeted subvolume, and direct addressing of nonorthogonal slices. The three-dimensional aspect of the data is further accommodated by thresholding and edge definition in space. Finally, morphological information, which is sparse in scintigraphic slices, is recaptured by the generation of planar data derived from data processed in the three-dimensional space.

A thresholding for radionuclide angiocardiography.

A thresholding algorithm for nuclear angiocardiographic studies is presented. The algorithm is free of operator intervention. The threshold is defined on the basis of the distribution of picture elements with decreasing count rates during systole. The application is valid for first-pass and equilibrium ECG gated studies. Reproducibility by different operators with various degrees of experience is high; concordance of the results with those of contrast studies is within the published ranges. The method provides a reliable step in full automation and standardization of scintigraphic angiocardiographies.

A fully automated determination of the left ventricular region of interest in nuclear angiocardiography.

The precise delineation of the left ventricular projection area is an essential part in the quantitative analysis of nuclear angiocardiograms. We have devised an algorithm that permits automation of this step, based on a one-dimensional Laplace operator whose kernel is 2, 2, -2, -4, -2, 2, 2. The operator characteristically enhances "valleys" more than edges and, therefore, favors septal and the valve plane detection. The operator is applied vertically, horizontally, and along both diagonals. Each pass is immediately followed by a local maximum search during which the image resulting from the Laplacian operator is reduced to a binary one, with zeros everywhere except where a local maximum was found along the path of the operator. This resultant image yields a closed "edge" around the left ventricle, even though many structures outside the left ventricle are also delineated. However, the centroid of the ventricle is defined from functional criteria and the region of interest is defined from centroid to first edge. The method has been applied to first-pass and gated studies in anterior and 45 degree left anterior oblique views. In 100 successive cases the ejection fraction obtained automatically was compared to the manual result. The regression equation yielded the relation: automatic method (%) = 1.7 + 1.0 manual method (%) +/-2% (r = 0.995), which is not significantly different from the identity relation. The failure rate was low (13%) but varied from 28% in the first-pass studies in the anterior view, to less than 8% in gated studies in the left ventricular oblique projection.